Schmidhuber: banning superintelligence is infeasible as compute gets 10x cheaper every 5 years
SchmidhuberAI · x · 2026-09-24
Replying to Bernie Sanders' call to ban superintelligence, Jürgen Schmidhuber argues this is infeasible: compute gets 10x cheaper every 5 years, and soon millions will run frontier labs and recursive self-improvement algorithms on desktops. In a technical note (IDSIA-9-26), he traces his own RSI lineage — first concrete RSI algorithms in 1987 (when compute was 100,000,000x more expensive), self-modifying policies since 1994, gradient-based RSI since 1992, and the Gödel Machine since 2003 — claiming RSI now drives both software and physical AI.
More from AGI Musings
- Beff Jezos mocks AI doomers: 'no real job, full-time doommongering' — mjdramstead · 2026-09-24
- Ex-OpenAI researcher joins debate claiming AI labs steal your prompts — suchenzang · 2026-09-24
- AI boom's 'teaser period': trillions in compute commitments come due in 2027-2028, article warns — sebpaquet · 2026-09-24
- AI beats forecasts: top AI ARR hits $100B vs $16-25B predicted, training runs cost $388M — koltregaskes · 2026-09-24
- When Everyone Has a Tireless AI Optimizer, the World Gets Far More Adversarial — RichmanRonald · 2026-09-24
- Noah Smith: Facebook's IPO and post-2016 internet drove real entrepreneurs away — MarvinTBaumann · 2026-09-24